Compare commits

..

1 Commits

Author SHA1 Message Date
Andrew Kane
0d8cc82908 Added HnswQuery struct [skip ci] 2024-02-04 22:30:59 -08:00
45 changed files with 218 additions and 1839 deletions

View File

@@ -28,7 +28,7 @@ jobs:
dev-files: true
- run: make
env:
PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
- run: |
export PG_CONFIG=`which pg_config`
sudo --preserve-env=PG_CONFIG make install
@@ -49,7 +49,7 @@ jobs:
postgres-version: 14
- run: make
env:
PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter
PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter
- run: make install
- run: make installcheck
- if: ${{ failure() }}
@@ -60,9 +60,7 @@ jobs:
wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_10.tar.gz
tar xf REL_14_10.tar.gz
- run: make prove_installcheck PROVE_FLAGS="-I ./postgres-REL_14_10/src/test/perl" PERL5LIB="/Users/runner/perl5/lib/perl5"
- run: make clean && /usr/local/opt/llvm@15/bin/scan-build --status-bugs make
env:
PG_CFLAGS: -DUSE_ASSERT_CHECKING
- run: make clean && /usr/local/opt/llvm@15/bin/scan-build --status-bugs make PG_CFLAGS="-DUSE_ASSERT_CHECKING"
windows:
runs-on: windows-latest
if: ${{ !startsWith(github.ref_name, 'mac') }}
@@ -73,7 +71,6 @@ jobs:
postgres-version: 14
- run: |
call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
cd %TEMP% && ^
nmake /NOLOGO /F Makefile.win && ^
nmake /NOLOGO /F Makefile.win install && ^
nmake /NOLOGO /F Makefile.win installcheck && ^
@@ -100,15 +97,4 @@ jobs:
sudo -u postgres make installcheck
sudo -u postgres make prove_installcheck
env:
PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
valgrind:
if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: ankane/setup-postgres-valgrind@v1
with:
postgres-version: 16
- run: make
- run: sudo --preserve-env=PG_CONFIG make install
- run: make installcheck
PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare

View File

@@ -1,16 +1,6 @@
## 0.7.0 (unreleased)
- Added `intvec` type
## 0.6.2 (2024-03-18)
- Reduced lock contention with parallel HNSW index builds
## 0.6.1 (2024-03-04)
## 0.6.1 (unreleased)
- Fixed error with `ANALYZE` and vectors with different dimensions
- Fixed segmentation fault with `shared_preload_libraries`
- Fixed vector subtraction being marked as commutative
## 0.6.0 (2024-01-29)

View File

@@ -1,4 +1,4 @@
Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
Portions Copyright (c) 1994, The Regents of the University of California

View File

@@ -2,7 +2,7 @@
"name": "vector",
"abstract": "Open-source vector similarity search for Postgres",
"description": "Supports L2 distance, inner product, and cosine distance",
"version": "0.6.2",
"version": "0.6.0",
"maintainer": [
"Andrew Kane <andrew@ankane.org>"
],
@@ -20,7 +20,7 @@
"vector": {
"file": "sql/vector.sql",
"docfile": "README.md",
"version": "0.6.2",
"version": "0.6.0",
"abstract": "Open-source vector similarity search for Postgres"
}
},

View File

@@ -1,10 +1,10 @@
EXTENSION = vector
EXTVERSION = 0.6.2
EXTVERSION = 0.6.0
MODULE_big = vector
DATA = $(wildcard sql/*--*.sql)
OBJS = src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/intvec.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/vector.o
HEADERS = src/intvec.h src/vector.h
OBJS = src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/vector.o
HEADERS = src/vector.h
TESTS = $(wildcard test/sql/*.sql)
REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
@@ -71,9 +71,11 @@ PG_MAJOR ?= 16
.PHONY: docker
docker:
docker build --pull --no-cache --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
docker build --pull --no-cache --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) .
docker build --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
.PHONY: docker-release
docker-release:
docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) .
docker buildx build --push --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .

View File

@@ -1,8 +1,8 @@
EXTENSION = vector
EXTVERSION = 0.6.2
EXTVERSION = 0.6.0
OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\intvec.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
HEADERS = src\intvec.h src\vector.h
OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
HEADERS = src\vector.h
REGRESS = btree cast copy functions input ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_options ivfflat_unlogged
REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)

208
README.md
View File

@@ -20,15 +20,15 @@ Compile and install the extension (supports Postgres 12+)
```sh
cd /tmp
git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
cd pgvector
make
make install # may need sudo
```
See the [installation notes](#installation-notes---linux-and-mac) if you run into issues
See the [installation notes](#installation-notes) if you run into issues
You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), [pkg](#pkg), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres). There are also instructions for [GitHub Actions](https://github.com/pgvector/setup-pgvector).
You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres). There are also instructions for [GitHub Actions](https://github.com/pgvector/setup-pgvector).
### Windows
@@ -44,15 +44,12 @@ Then use `nmake` to build:
```cmd
set "PGROOT=C:\Program Files\PostgreSQL\16"
cd %TEMP%
git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
cd pgvector
nmake /F Makefile.win
nmake /F Makefile.win install
```
See the [installation notes](#installation-notes---windows) if you run into issues
You can also install it with [Docker](#docker) or [conda-forge](#conda-forge).
## Getting Started
@@ -105,12 +102,6 @@ Insert vectors
INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
```
Or load vectors in bulk using `COPY` ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/bulk_loading.py))
```sql
COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
```
Upsert vectors
```sql
@@ -273,8 +264,6 @@ HINT: Increase maintenance_work_mem to speed up builds.
Note: Do not set `maintenance_work_mem` so high that it exhausts the memory on the server
Like other index types, its faster to create an index after loading your initial data
Starting with 0.6.0, you can also speed up index creation by increasing the number of parallel workers (2 by default)
```sql
@@ -413,51 +402,13 @@ You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python
## Performance
### Tuning
Use a tool like [PgTune](https://pgtune.leopard.in.ua/) to set initial values for Postgres server parameters. For instance, `shared_buffers` should typically be 25% of the servers memory. You can find the config file with:
```sql
SHOW config_file;
```
And check individual settings with:
```sql
SHOW shared_buffers;
```
Be sure to restart Postgres for changes to take effect.
### Loading
Use `COPY` for bulk loading data ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/bulk_loading.py)).
```sql
COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
```
Add any indexes *after* loading the initial data for best performance.
### Indexing
See index build time for [HNSW](#index-build-time) and [IVFFlat](#index-build-time-1).
In production environments, create indexes concurrently to avoid blocking writes.
```sql
CREATE INDEX CONCURRENTLY ...
```
### Querying
Use `EXPLAIN ANALYZE` to debug performance.
```sql
EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
```
#### Exact Search
### Exact Search
To speed up queries without an index, increase `max_parallel_workers_per_gather`.
@@ -471,7 +422,7 @@ If vectors are normalized to length 1 (like [OpenAI embeddings](https://platform
SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
```
#### Approximate Search
### Approximate Search
To speed up queries with an IVFFlat index, increase the number of inverted lists (at the expense of recall).
@@ -479,7 +430,7 @@ To speed up queries with an IVFFlat index, increase the number of inverted lists
CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 1000);
```
### Vacuuming
## Vacuuming
Vacuuming can take a while for HNSW indexes. Speed it up by reindexing first.
@@ -488,41 +439,6 @@ REINDEX INDEX CONCURRENTLY index_name;
VACUUM table_name;
```
## Monitoring
Monitor performance with [pg_stat_statements](https://www.postgresql.org/docs/current/pgstatstatements.html) (be sure to add it to `shared_preload_libraries`).
```sql
CREATE EXTENSION pg_stat_statements;
```
Get the most time-consuming queries with:
```sql
SELECT query, calls, ROUND((total_plan_time + total_exec_time) / calls) AS avg_time_ms,
ROUND((total_plan_time + total_exec_time) / 60000) AS total_time_min
FROM pg_stat_statements ORDER BY total_plan_time + total_exec_time DESC LIMIT 20;
```
Note: Replace `total_plan_time + total_exec_time` with `total_time` for Postgres < 13
Monitor recall by comparing results from approximate search with exact search.
```sql
BEGIN;
SET LOCAL enable_indexscan = off; -- use exact search
SELECT ...
COMMIT;
```
## Scaling
Scale pgvector the same way you scale Postgres.
Scale vertically by increasing memory, CPU, and storage on a single instance. Use existing tools to [tune parameters](#tuning) and [monitor performance](#monitoring).
Scale horizontally with [replicas](https://www.postgresql.org/docs/current/hot-standby.html), or use [Citus](https://github.com/citusdata/citus) or another approach for sharding ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/citus.py)).
## Languages
Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
@@ -616,18 +532,6 @@ and query with:
SELECT * FROM items ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
```
#### Are binary vectors supported?
You can store binary vectors and perform exact nearest neighbor search by Hamming distance in Postgres without an extension ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/hash_image_search.py)).
```tsql
CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
INSERT INTO items (embedding) VALUES (B'000'), (B'111');
SELECT * FROM items ORDER BY bit_count(embedding # B'101') LIMIT 5;
```
Indexing is not currently supported.
#### Do indexes need to fit into memory?
No, but like other index types, youll likely see better performance if they do. You can get the size of an index with:
@@ -640,17 +544,7 @@ SELECT pg_size_pretty(pg_relation_size('index_name'));
#### Why isnt a query using an index?
The query needs to have an `ORDER BY` and `LIMIT`, and the `ORDER BY` must be the result of a distance operator, not an expression.
```sql
-- index
ORDER BY embedding <=> '[3,1,2]' LIMIT 5;
-- no index
ORDER BY 1 - (embedding <=> '[3,1,2]') DESC LIMIT 5;
```
You can encourage the planner to use an index for a query with:
The cost estimation in pgvector < 0.4.3 does not always work well with the planner. You can encourage the planner to use an index for a query with:
```sql
BEGIN;
@@ -679,12 +573,6 @@ or choose to store vectors inline:
ALTER TABLE items ALTER COLUMN embedding SET STORAGE PLAIN;
```
#### Why are there less results for a query after adding an HNSW index?
Results are limited by the size of the dynamic candidate list (`hnsw.ef_search`). There may be even less results due to dead tuples or filtering conditions in the query. We recommend setting `hnsw.ef_search` to at least twice the `LIMIT` of the query. If you need more than 500 results, use an IVFFlat index instead.
Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
#### Why are there less results for a query after adding an IVFFlat index?
The index was likely created with too little data for the number of lists. Drop the index until the table has more data.
@@ -693,15 +581,11 @@ The index was likely created with too little data for the number of lists. Drop
DROP INDEX index_name;
```
Results can also be limited by the number of probes (`ivfflat.probes`).
Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
## Reference
### Vector Type
Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single-precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
### Vector Operators
@@ -725,35 +609,14 @@ l1_distance(vector, vector) → double precision | taxicab distance | 0.5.0
vector_dims(vector) → integer | number of dimensions |
vector_norm(vector) → double precision | Euclidean norm |
### Vector Aggregate Functions
### Aggregate Functions
Function | Description | Added
--- | --- | ---
avg(vector) → vector | average |
sum(vector) → vector | sum | 0.5.0
### Intvec Type
Each int vector takes `dimensions + 8` bytes of storage. Each element is a single byte signed integer. Int vectors can have up to 16,000 dimensions.
### Intvec Operators
Operator | Description | Added
--- | --- | ---
<-> | Euclidean distance | 0.7.0
<#> | negative inner product | 0.7.0
<=> | cosine distance | 0.7.0
### Intvec Functions
Function | Description | Added
--- | --- | ---
cosine_distance(intvec, intvec) → double precision | cosine distance | 0.7.0
inner_product(intvec, intvec) → double precision | inner product | 0.7.0
l2_distance(intvec, intvec) → double precision | Euclidean distance | 0.7.0
l1_distance(intvec, intvec) → double precision | taxicab distance | 0.7.0
## Installation Notes - Linux and Mac
## Installation Notes
### Postgres Location
@@ -793,26 +656,6 @@ Note: Replace `16` with your Postgres server version
If compilation fails and the output includes `warning: no such sysroot directory` on Mac, reinstall Xcode Command Line Tools.
### Portability
By default, pgvector compiles with `-march=native` on some platforms for best performance. However, this can lead to `Illegal instruction` errors if trying to run the compiled extension on a different machine.
To compile for portability, use:
```sh
make OPTFLAGS=""
```
## Installation Notes - Windows
### Missing Header
If compilation fails with `Cannot open include file: 'postgres.h': No such file or directory`, make sure `PGROOT` is correct.
### Permissions
If installation fails with `Access is denied`, re-run the installation instructions as an administrator.
## Additional Installation Methods
### Docker
@@ -828,7 +671,7 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
You can also build the image manually:
```sh
git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
cd pgvector
docker build --build-arg PG_MAJOR=16 -t myuser/pgvector .
```
@@ -873,21 +716,6 @@ sudo dnf install pgvector_16
Note: Replace `16` with your Postgres server version
### pkg
Install the FreeBSD package with:
```sh
pkg install postgresql15-pg_vector
```
or the port with:
```sh
cd /usr/ports/databases/pgvector
make install
```
### conda-forge
With Conda Postgres, install from [conda-forge](https://anaconda.org/conda-forge/pgvector) with:
@@ -997,12 +825,6 @@ make installcheck REGRESS=functions # regression test
make prove_installcheck PROVE_TESTS=test/t/001_ivfflat_wal.pl # TAP test
```
To enable assertions:
```sh
make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
```
To enable benchmarking:
```sh
@@ -1015,6 +837,12 @@ To show memory usage:
make clean && PG_CFLAGS="-DHNSW_MEMORY -DIVFFLAT_MEMORY" make && make install
```
To enable assertions:
```sh
make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
```
To get k-means metrics:
```sh

View File

@@ -1,16 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.1'" to load this file. \quit
DROP OPERATOR - (vector, vector);
CREATE OPERATOR - (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub
);
ALTER OPERATOR <= (vector, vector) SET (
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
);
ALTER OPERATOR >= (vector, vector) SET (
RESTRICT = scalargesel, JOIN = scalargejoinsel
);

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.2'" to load this file. \quit

View File

@@ -1,92 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.0'" to load this file. \quit
CREATE TYPE intvec;
CREATE FUNCTION intvec_in(cstring, oid, integer) RETURNS intvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_out(intvec) RETURNS cstring
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_typmod_in(cstring[]) RETURNS integer
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_recv(internal, oid, integer) RETURNS intvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_send(intvec) RETURNS bytea
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE TYPE intvec (
INPUT = intvec_in,
OUTPUT = intvec_out,
TYPMOD_IN = intvec_typmod_in,
RECEIVE = intvec_recv,
SEND = intvec_send,
STORAGE = external
);
CREATE FUNCTION l2_distance(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION inner_product(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION cosine_distance(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l1_distance(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_norm(intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_l2_squared_distance(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_negative_inner_product(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec(intvec, integer, boolean) RETURNS intvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_intvec(integer[], integer, boolean) RETURNS intvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE CAST (intvec AS intvec)
WITH FUNCTION intvec(intvec, integer, boolean) AS IMPLICIT;
CREATE CAST (integer[] AS intvec)
WITH FUNCTION array_to_intvec(integer[], integer, boolean) AS ASSIGNMENT;
CREATE OPERATOR <-> (
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = l2_distance,
COMMUTATOR = '<->'
);
CREATE OPERATOR <#> (
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_negative_inner_product,
COMMUTATOR = '<#>'
);
CREATE OPERATOR <=> (
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = cosine_distance,
COMMUTATOR = '<=>'
);
CREATE OPERATOR CLASS intvec_l2_ops
FOR TYPE intvec USING hnsw AS
OPERATOR 1 <-> (intvec, intvec) FOR ORDER BY float_ops,
FUNCTION 1 intvec_l2_squared_distance(intvec, intvec);
CREATE OPERATOR CLASS intvec_ip_ops
FOR TYPE intvec USING hnsw AS
OPERATOR 1 <#> (intvec, intvec) FOR ORDER BY float_ops,
FUNCTION 1 intvec_negative_inner_product(intvec, intvec);
CREATE OPERATOR CLASS intvec_cosine_ops
FOR TYPE intvec USING hnsw AS
OPERATOR 1 <=> (intvec, intvec) FOR ORDER BY float_ops,
FUNCTION 1 cosine_distance(intvec, intvec),
FUNCTION 2 l2_norm(intvec);

View File

@@ -180,7 +180,8 @@ CREATE OPERATOR + (
);
CREATE OPERATOR - (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub,
COMMUTATOR = -
);
CREATE OPERATOR * (
@@ -194,10 +195,11 @@ CREATE OPERATOR < (
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
);
-- should use scalarlesel and scalarlejoinsel, but not supported in Postgres < 11
CREATE OPERATOR <= (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_le,
COMMUTATOR = >= , NEGATOR = > ,
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
);
CREATE OPERATOR = (
@@ -212,10 +214,11 @@ CREATE OPERATOR <> (
RESTRICT = eqsel, JOIN = eqjoinsel
);
-- should use scalargesel and scalargejoinsel, but not supported in Postgres < 11
CREATE OPERATOR >= (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_ge,
COMMUTATOR = <= , NEGATOR = < ,
RESTRICT = scalargesel, JOIN = scalargejoinsel
RESTRICT = scalargtsel, JOIN = scalargtjoinsel
);
CREATE OPERATOR > (
@@ -287,107 +290,3 @@ CREATE OPERATOR CLASS vector_cosine_ops
OPERATOR 1 <=> (vector, vector) FOR ORDER BY float_ops,
FUNCTION 1 vector_negative_inner_product(vector, vector),
FUNCTION 2 vector_norm(vector);
-- intvec type
CREATE TYPE intvec;
CREATE FUNCTION intvec_in(cstring, oid, integer) RETURNS intvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_out(intvec) RETURNS cstring
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_typmod_in(cstring[]) RETURNS integer
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_recv(internal, oid, integer) RETURNS intvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_send(intvec) RETURNS bytea
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE TYPE intvec (
INPUT = intvec_in,
OUTPUT = intvec_out,
TYPMOD_IN = intvec_typmod_in,
RECEIVE = intvec_recv,
SEND = intvec_send,
STORAGE = external
);
-- intvec functions
CREATE FUNCTION l2_distance(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION inner_product(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION cosine_distance(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l1_distance(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_norm(intvec) RETURNS float8
AS 'MODULE_PATHNAME', 'intvec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- intvec private functions
CREATE FUNCTION intvec_l2_squared_distance(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION intvec_negative_inner_product(intvec, intvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- intvec cast functions
CREATE FUNCTION intvec(intvec, integer, boolean) RETURNS intvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_intvec(integer[], integer, boolean) RETURNS intvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- intvec casts
CREATE CAST (intvec AS intvec)
WITH FUNCTION intvec(intvec, integer, boolean) AS IMPLICIT;
CREATE CAST (integer[] AS intvec)
WITH FUNCTION array_to_intvec(integer[], integer, boolean) AS ASSIGNMENT;
-- intvec operators
CREATE OPERATOR <-> (
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = l2_distance,
COMMUTATOR = '<->'
);
CREATE OPERATOR <#> (
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_negative_inner_product,
COMMUTATOR = '<#>'
);
CREATE OPERATOR <=> (
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = cosine_distance,
COMMUTATOR = '<=>'
);
-- intvec opclasses
CREATE OPERATOR CLASS intvec_l2_ops
FOR TYPE intvec USING hnsw AS
OPERATOR 1 <-> (intvec, intvec) FOR ORDER BY float_ops,
FUNCTION 1 intvec_l2_squared_distance(intvec, intvec);
CREATE OPERATOR CLASS intvec_ip_ops
FOR TYPE intvec USING hnsw AS
OPERATOR 1 <#> (intvec, intvec) FOR ORDER BY float_ops,
FUNCTION 1 intvec_negative_inner_product(intvec, intvec);
CREATE OPERATOR CLASS intvec_cosine_ops
FOR TYPE intvec USING hnsw AS
OPERATOR 1 <=> (intvec, intvec) FOR ORDER BY float_ops,
FUNCTION 1 cosine_distance(intvec, intvec),
FUNCTION 2 l2_norm(intvec);

View File

@@ -8,7 +8,6 @@
#include "commands/progress.h"
#include "commands/vacuum.h"
#include "hnsw.h"
#include "miscadmin.h"
#include "utils/guc.h"
#include "utils/selfuncs.h"
@@ -29,7 +28,7 @@ static relopt_kind hnsw_relopt_kind;
* this grows bigger, we should use a shmem_request_hook and
* RequestAddinShmemSpace() to pre-reserve space for this.
*/
void
static void
HnswInitLockTranche(void)
{
int *tranche_ids;
@@ -54,8 +53,7 @@ HnswInitLockTranche(void)
void
HnswInit(void)
{
if (!process_shared_preload_libraries_in_progress)
HnswInitLockTranche();
HnswInitLockTranche();
hnsw_relopt_kind = add_reloption_kind();
add_int_reloption(hnsw_relopt_kind, "m", "Max number of connections",

View File

@@ -55,12 +55,6 @@
#define HNSW_UPDATE_ENTRY_GREATER 1
#define HNSW_UPDATE_ENTRY_ALWAYS 2
typedef enum HnswType
{
HNSW_TYPE_VECTOR,
HNSW_TYPE_INTVEC
} HnswType;
/* Build phases */
/* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
#define PROGRESS_HNSW_PHASE_LOAD 2
@@ -86,7 +80,7 @@ typedef enum HnswType
#if PG_VERSION_NUM < 130000
#define list_delete_last(list) list_truncate(list, list_length(list) - 1)
#define list_sort(list, cmp) ((list) = list_qsort(list, cmp))
#define list_sort(list, cmp) list_qsort(list, cmp)
#endif
#define HnswIsElementTuple(tup) ((tup)->type == HNSW_ELEMENT_TUPLE_TYPE)
@@ -135,7 +129,7 @@ HnswPtrDeclare(HnswNeighborArray, HnswNeighborArrayRelptr, HnswNeighborArrayPtr)
HnswPtrDeclare(HnswNeighborArrayPtr, HnswNeighborsRelptr, HnswNeighborsPtr);
HnswPtrDeclare(char, DatumRelptr, DatumPtr);
struct HnswElementData
typedef struct HnswElementData
{
HnswElementPtr next;
ItemPointerData heaptids[HNSW_HEAPTIDS];
@@ -150,7 +144,7 @@ struct HnswElementData
BlockNumber neighborPage;
DatumPtr value;
LWLock lock;
};
} HnswElementData;
typedef HnswElementData * HnswElement;
@@ -161,12 +155,12 @@ typedef struct HnswCandidate
bool closer;
} HnswCandidate;
struct HnswNeighborArray
typedef struct HnswNeighborArray
{
int length;
bool closerSet;
HnswCandidate items[FLEXIBLE_ARRAY_MEMBER];
};
} HnswNeighborArray;
typedef struct HnswPairingHeapNode
{
@@ -191,7 +185,6 @@ typedef struct HnswGraph
/* Entry state */
LWLock entryLock;
LWLock entryWaitLock;
HnswElementPtr entryPoint;
/* Allocations state */
@@ -248,7 +241,6 @@ typedef struct HnswBuildState
Relation index;
IndexInfo *indexInfo;
ForkNumber forkNum;
HnswType type;
/* Settings */
int dimensions;
@@ -269,6 +261,7 @@ typedef struct HnswBuildState
HnswGraph *graph;
double ml;
int maxLevel;
Vector *normvec;
/* Memory */
MemoryContext graphCtx;
@@ -369,23 +362,27 @@ typedef struct HnswVacuumState
MemoryContext tmpCtx;
} HnswVacuumState;
typedef struct HnswQuery
{
Datum value;
} HnswQuery;
/* Methods */
int HnswGetM(Relation index);
int HnswGetEfConstruction(Relation index);
FmgrInfo *HnswOptionalProcInfo(Relation index, uint16 procnum);
HnswType HnswGetType(Relation index);
bool HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type);
bool HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
Buffer HnswNewBuffer(Relation index, ForkNumber forkNum);
void HnswInitPage(Buffer buf, Page page);
void HnswInit(void);
List *HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement);
List *HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement);
HnswElement HnswGetEntryPoint(Relation index);
void HnswGetMetaPageInfo(Relation index, int *m, HnswElement * entryPoint);
void *HnswAlloc(HnswAllocator * allocator, Size size);
HnswElement HnswInitElement(char *base, ItemPointer tid, int m, double ml, int maxLevel, HnswAllocator * alloc);
HnswElement HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno);
void HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, bool existing);
HnswCandidate *HnswEntryCandidate(char *base, HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
HnswCandidate *HnswEntryCandidate(char *base, HnswElement em, HnswQuery * q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
void HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum, bool building);
void HnswSetNeighborTuple(char *base, HnswNeighborTuple ntup, HnswElement e, int m);
void HnswAddHeapTid(HnswElement element, ItemPointer heaptid);
@@ -393,11 +390,10 @@ void HnswInitNeighbors(char *base, HnswElement element, int m, HnswAllocator *
bool HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull, ItemPointer heap_tid, bool building);
void HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building);
void HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec);
void HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec);
void HnswLoadElement(HnswElement element, float *distance, HnswQuery * q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec);
void HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element);
void HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation);
void HnswLoadNeighbors(HnswElement element, Relation index, int m);
void HnswInitLockTranche(void);
PGDLLEXPORT void HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc);
/* Index access methods */

View File

@@ -176,7 +176,7 @@ CreateGraphPages(HnswBuildState * buildstate)
Size etupSize;
Size ntupSize;
Size combinedSize;
Pointer valuePtr = HnswPtrAccess(base, element->value);
void *valuePtr = HnswPtrAccess(base, element->value);
/* Update iterator */
iter = element->next;
@@ -431,15 +431,10 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
HnswGraph *graph = buildstate->graph;
HnswElement entryPoint;
LWLock *entryLock = &graph->entryLock;
LWLock *entryWaitLock = &graph->entryWaitLock;
int efConstruction = buildstate->efConstruction;
int m = buildstate->m;
char *base = buildstate->hnswarea;
/* Wait if another process needs exclusive lock on entry lock */
LWLockAcquire(entryWaitLock, LW_EXCLUSIVE);
LWLockRelease(entryWaitLock);
/* Get entry point */
LWLockAcquire(entryLock, LW_SHARED);
entryPoint = HnswPtrAccess(base, graph->entryPoint);
@@ -450,10 +445,8 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
/* Release shared lock */
LWLockRelease(entryLock);
/* Tell other processes to wait and get exclusive lock */
LWLockAcquire(entryWaitLock, LW_EXCLUSIVE);
/* Get exclusive lock */
LWLockAcquire(entryLock, LW_EXCLUSIVE);
LWLockRelease(entryWaitLock);
/* Get latest entry point after lock is acquired */
entryPoint = HnswPtrAccess(base, graph->entryPoint);
@@ -489,7 +482,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
/* Normalize if needed */
if (buildstate->normprocinfo != NULL)
{
if (!HnswNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->type))
if (!HnswNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec))
return false;
}
@@ -608,9 +601,6 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
static void
InitGraph(HnswGraph * graph, char *base, long memoryTotal)
{
/* Initialize the lock tranche if needed */
HnswInitLockTranche();
HnswPtrStore(base, graph->head, (HnswElement) NULL);
HnswPtrStore(base, graph->entryPoint, (HnswElement) NULL);
graph->memoryUsed = 0;
@@ -619,7 +609,6 @@ InitGraph(HnswGraph * graph, char *base, long memoryTotal)
graph->indtuples = 0;
SpinLockInit(&graph->lock);
LWLockInitialize(&graph->entryLock, hnsw_lock_tranche_id);
LWLockInitialize(&graph->entryWaitLock, hnsw_lock_tranche_id);
LWLockInitialize(&graph->allocatorLock, hnsw_lock_tranche_id);
LWLockInitialize(&graph->flushLock, hnsw_lock_tranche_id);
}
@@ -671,27 +660,21 @@ HnswSharedMemoryAlloc(Size size, void *state)
static void
InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, IndexInfo *indexInfo, ForkNumber forkNum)
{
int maxDimensions = HNSW_MAX_DIM;
buildstate->heap = heap;
buildstate->index = index;
buildstate->indexInfo = indexInfo;
buildstate->forkNum = forkNum;
buildstate->type = HnswGetType(index);
buildstate->m = HnswGetM(index);
buildstate->efConstruction = HnswGetEfConstruction(index);
buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
if (buildstate->type == HNSW_TYPE_INTVEC)
maxDimensions *= 4;
/* Require column to have dimensions to be indexed */
if (buildstate->dimensions < 0)
elog(ERROR, "column does not have dimensions");
if (buildstate->dimensions > maxDimensions)
elog(ERROR, "column cannot have more than %d dimensions for hnsw index", maxDimensions);
if (buildstate->dimensions > HNSW_MAX_DIM)
elog(ERROR, "column cannot have more than %d dimensions for hnsw index", HNSW_MAX_DIM);
if (buildstate->efConstruction < 2 * buildstate->m)
elog(ERROR, "ef_construction must be greater than or equal to 2 * m");
@@ -709,6 +692,9 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
buildstate->ml = HnswGetMl(buildstate->m);
buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);
/* Reuse for each tuple */
buildstate->normvec = InitVector(buildstate->dimensions);
buildstate->graphCtx = GenerationContextCreate(CurrentMemoryContext,
"Hnsw build graph context",
#if PG_VERSION_NUM >= 150000
@@ -732,6 +718,7 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
static void
FreeBuildState(HnswBuildState * buildstate)
{
pfree(buildstate->normvec);
MemoryContextDelete(buildstate->graphCtx);
MemoryContextDelete(buildstate->tmpCtx);
}
@@ -987,14 +974,6 @@ HnswBeginParallel(HnswBuildState * buildstate, bool isconcurrent, int request)
/* Report less than allocated so never fails */
InitGraph(&hnswshared->graphData, hnswarea, esthnswarea - 1024 * 1024);
/*
* Avoid base address for relptr for Postgres < 14.5
* https://github.com/postgres/postgres/commit/7201cd18627afc64850537806da7f22150d1a83b
*/
#if PG_VERSION_NUM < 140005
hnswshared->graphData.memoryUsed += MAXALIGN(1);
#endif
shm_toc_insert(pcxt->toc, PARALLEL_KEY_HNSW_SHARED, hnswshared);
shm_toc_insert(pcxt->toc, PARALLEL_KEY_HNSW_AREA, hnswarea);

View File

@@ -622,7 +622,7 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
if (normprocinfo != NULL)
{
if (!HnswNormValue(normprocinfo, collation, &value, HnswGetType(index)))
if (!HnswNormValue(normprocinfo, collation, &value, NULL))
return;
}

View File

@@ -11,7 +11,7 @@
* Algorithm 5 from paper
*/
static List *
GetScanItems(IndexScanDesc scan, Datum q)
GetScanItems(IndexScanDesc scan, Datum value)
{
HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
Relation index = scan->indexRelation;
@@ -22,6 +22,9 @@ GetScanItems(IndexScanDesc scan, Datum q)
int m;
HnswElement entryPoint;
char *base = NULL;
HnswQuery q;
q.value = value;
/* Get m and entry point */
HnswGetMetaPageInfo(index, &m, &entryPoint);
@@ -29,15 +32,38 @@ GetScanItems(IndexScanDesc scan, Datum q)
if (entryPoint == NULL)
return NIL;
ep = list_make1(HnswEntryCandidate(base, entryPoint, q, index, procinfo, collation, false));
ep = list_make1(HnswEntryCandidate(base, entryPoint, &q, index, procinfo, collation, false));
for (int lc = entryPoint->level; lc >= 1; lc--)
{
w = HnswSearchLayer(base, q, ep, 1, lc, index, procinfo, collation, m, false, NULL);
w = HnswSearchLayer(base, &q, ep, 1, lc, index, procinfo, collation, m, false, NULL);
ep = w;
}
return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL);
return HnswSearchLayer(base, &q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL);
}
/*
* Get dimensions from metapage
*/
static int
GetDimensions(Relation index)
{
Buffer buf;
Page page;
HnswMetaPage metap;
int dimensions;
buf = ReadBuffer(index, HNSW_METAPAGE_BLKNO);
LockBuffer(buf, BUFFER_LOCK_SHARE);
page = BufferGetPage(buf);
metap = HnswPageGetMeta(page);
dimensions = metap->dimensions;
UnlockReleaseBuffer(buf);
return dimensions;
}
/*
@@ -50,7 +76,7 @@ GetScanValue(IndexScanDesc scan)
Datum value;
if (scan->orderByData->sk_flags & SK_ISNULL)
value = PointerGetDatum(NULL);
value = PointerGetDatum(InitVector(GetDimensions(scan->indexRelation)));
else
{
value = scan->orderByData->sk_argument;
@@ -61,7 +87,7 @@ GetScanValue(IndexScanDesc scan)
/* Fine if normalization fails */
if (so->normprocinfo != NULL)
HnswNormValue(so->normprocinfo, so->collation, &value, HnswGetType(scan->indexRelation));
HnswNormValue(so->normprocinfo, so->collation, &value, NULL);
}
return value;

View File

@@ -1,17 +1,14 @@
#include "postgres.h"
#include <float.h>
#include <math.h>
#include "access/generic_xlog.h"
#include "catalog/pg_type.h"
#include "hnsw.h"
#include "lib/pairingheap.h"
#include "storage/bufmgr.h"
#include "utils/datum.h"
#include "utils/memdebug.h"
#include "utils/rel.h"
#include "utils/syscache.h"
#include "vector.h"
#if PG_VERSION_NUM >= 130000
@@ -152,32 +149,6 @@ HnswOptionalProcInfo(Relation index, uint16 procnum)
return index_getprocinfo(index, 1, procnum);
}
/*
* Get vector type
*/
HnswType
HnswGetType(Relation index)
{
Oid typeOid = TupleDescAttr(index->rd_att, 0)->atttypid;
HeapTuple tuple;
Form_pg_type type;
int result;
tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(typeOid));
if (!HeapTupleIsValid(tuple))
elog(ERROR, "cache lookup failed for type %u", typeOid);
type = (Form_pg_type) GETSTRUCT(tuple);
if (strcmp(NameStr(type->typname), "intvec") == 0)
result = HNSW_TYPE_INTVEC;
else
result = HNSW_TYPE_VECTOR;
ReleaseSysCache(tuple);
return result;
}
/*
* Divide by the norm
*
@@ -187,29 +158,21 @@ HnswGetType(Relation index)
* if it's different than the original value
*/
bool
HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type)
HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
{
double norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
if (norm > 0)
{
/* TODO Remove vector-specific code */
if (type == HNSW_TYPE_VECTOR)
{
Vector *v = DatumGetVector(*value);
Vector *result = InitVector(v->dim);
Vector *v = DatumGetVector(*value);
for (int i = 0; i < v->dim; i++)
result->x[i] = v->x[i] / norm;
if (result == NULL)
result = InitVector(v->dim);
*value = PointerGetDatum(result);
}
else if (type == HNSW_TYPE_INTVEC)
{
/* Do nothing */
}
else
elog(ERROR, "Unsupported type");
for (int i = 0; i < v->dim; i++)
result->x[i] = v->x[i] / norm;
*value = PointerGetDatum(result);
return true;
}
@@ -592,7 +555,7 @@ HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHe
* Load an element and optionally get its distance from q
*/
void
HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
HnswLoadElement(HnswElement element, float *distance, HnswQuery * q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
{
Buffer buf;
Page page;
@@ -612,19 +575,7 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
/* Calculate distance */
if (distance != NULL)
{
if (DatumGetPointer(*q) == NULL)
*distance = 0;
else
{
*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->data)));
/* Needed for intvec cosine distance */
/* TODO Improve */
if (isnan(*distance))
*distance = FLT_MAX;
}
}
*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, q->value, PointerGetDatum(&etup->data)));
UnlockReleaseBuffer(buf);
}
@@ -633,19 +584,19 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
* Get the distance for a candidate
*/
static float
GetCandidateDistance(char *base, HnswCandidate * hc, Datum q, FmgrInfo *procinfo, Oid collation)
GetCandidateDistance(char *base, HnswCandidate * hc, HnswQuery * q, FmgrInfo *procinfo, Oid collation)
{
HnswElement hce = HnswPtrAccess(base, hc->element);
Datum value = HnswGetValue(base, hce);
return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, q, value));
return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, q->value, value));
}
/*
* Create a candidate for the entry point
*/
HnswCandidate *
HnswEntryCandidate(char *base, HnswElement entryPoint, Datum q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
HnswEntryCandidate(char *base, HnswElement entryPoint, HnswQuery * q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
{
HnswCandidate *hc = palloc(sizeof(HnswCandidate));
@@ -653,7 +604,7 @@ HnswEntryCandidate(char *base, HnswElement entryPoint, Datum q, Relation index,
if (index == NULL)
hc->distance = GetCandidateDistance(base, hc, q, procinfo, collation);
else
HnswLoadElement(entryPoint, &hc->distance, &q, index, procinfo, collation, loadVec);
HnswLoadElement(entryPoint, &hc->distance, q, index, procinfo, collation, loadVec);
return hc;
}
@@ -771,7 +722,7 @@ CountElement(char *base, HnswElement skipElement, HnswCandidate * hc)
* Algorithm 2 from paper
*/
List *
HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement)
HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement)
{
List *w = NIL;
pairingheap *C = pairingheap_allocate(CompareNearestCandidates, NULL);
@@ -855,7 +806,7 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
if (index == NULL)
eDistance = GetCandidateDistance(base, e, q, procinfo, collation);
else
HnswLoadElement(eElement, &eDistance, &q, index, procinfo, collation, inserting);
HnswLoadElement(eElement, &eDistance, q, index, procinfo, collation, inserting);
Assert(!eElement->deleted);
@@ -909,15 +860,12 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
static int
#if PG_VERSION_NUM >= 130000
CompareCandidateDistances(const ListCell *a, const ListCell *b)
{
HnswCandidate *hca = lfirst(a);
HnswCandidate *hcb = lfirst(b);
#else
CompareCandidateDistances(const void *a, const void *b)
{
HnswCandidate *hca = lfirst(*(ListCell **) a);
HnswCandidate *hcb = lfirst(*(ListCell **) b);
#endif
{
HnswCandidate *hca = lfirst((ListCell *) a);
HnswCandidate *hcb = lfirst((ListCell *) b);
if (hca->distance < hcb->distance)
return 1;
@@ -940,15 +888,12 @@ CompareCandidateDistances(const void *a, const void *b)
static int
#if PG_VERSION_NUM >= 130000
CompareCandidateDistancesOffset(const ListCell *a, const ListCell *b)
{
HnswCandidate *hca = lfirst(a);
HnswCandidate *hcb = lfirst(b);
#else
CompareCandidateDistancesOffset(const void *a, const void *b)
{
HnswCandidate *hca = lfirst(*(ListCell **) a);
HnswCandidate *hcb = lfirst(*(ListCell **) b);
#endif
{
HnswCandidate *hca = lfirst((ListCell *) a);
HnswCandidate *hcb = lfirst((ListCell *) b);
if (hca->distance < hcb->distance)
return 1;
@@ -1007,9 +952,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
{
List *r = NIL;
List *w = list_copy(c);
HnswCandidate **wd;
int wdlen = 0;
int wdoff = 0;
pairingheap *wd;
HnswNeighborArray *neighbors = HnswGetNeighbors(base, e2, lc);
bool mustCalculate = !neighbors->closerSet;
List *added = NIL;
@@ -1018,7 +961,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
if (list_length(w) <= lm)
return w;
wd = palloc(sizeof(HnswCandidate *) * list_length(w));
wd = pairingheap_allocate(CompareNearestCandidates, NULL);
/* Ensure order of candidates is deterministic for closer caching */
if (sortCandidates)
@@ -1084,21 +1027,21 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
if (e->closer)
r = lappend(r, e);
else
wd[wdlen++] = e;
pairingheap_add(wd, &(CreatePairingHeapNode(e)->ph_node));
}
/* Cached value can only be used in future if sorted deterministically */
neighbors->closerSet = sortCandidates;
/* Keep pruned connections */
while (wdoff < wdlen && list_length(r) < lm)
r = lappend(r, wd[wdoff++]);
while (!pairingheap_is_empty(wd) && list_length(r) < lm)
r = lappend(r, ((HnswPairingHeapNode *) pairingheap_remove_first(wd))->inner);
/* Return pruned for update connections */
if (pruned != NULL)
{
if (wdoff < wdlen)
*pruned = wd[wdoff];
if (!pairingheap_is_empty(wd))
*pruned = ((HnswPairingHeapNode *) pairingheap_first(wd))->inner;
else
*pruned = linitial(w);
}
@@ -1148,7 +1091,9 @@ HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm
/* Load elements on insert */
if (index != NULL)
{
Datum q = HnswGetValue(base, hce);
HnswQuery q;
q.value = HnswGetValue(base, hce);
for (int i = 0; i < currentNeighbors->length; i++)
{
@@ -1158,7 +1103,7 @@ HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm
if (HnswPtrIsNull(base, hc3Element->value))
HnswLoadElement(hc3Element, &hc3->distance, &q, index, procinfo, collation, true);
else
hc3->distance = GetCandidateDistance(base, hc3, q, procinfo, collation);
hc3->distance = GetCandidateDistance(base, hc3, &q, procinfo, collation);
/* Prune element if being deleted */
if (hc3Element->heaptidsLength == 0)
@@ -1258,9 +1203,11 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
List *w;
int level = element->level;
int entryLevel;
Datum q = HnswGetValue(base, element);
HnswQuery q;
HnswElement skipElement = existing ? element : NULL;
q.value = HnswGetValue(base, element);
#if PG_VERSION_NUM >= 130000
/* Precompute hash */
if (index == NULL)
@@ -1272,13 +1219,13 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
return;
/* Get entry point and level */
ep = list_make1(HnswEntryCandidate(base, entryPoint, q, index, procinfo, collation, true));
ep = list_make1(HnswEntryCandidate(base, entryPoint, &q, index, procinfo, collation, true));
entryLevel = entryPoint->level;
/* 1st phase: greedy search to insert level */
for (int lc = entryLevel; lc >= level + 1; lc--)
{
w = HnswSearchLayer(base, q, ep, 1, lc, index, procinfo, collation, m, true, skipElement);
w = HnswSearchLayer(base, &q, ep, 1, lc, index, procinfo, collation, m, true, skipElement);
ep = w;
}
@@ -1296,7 +1243,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
List *neighbors;
List *lw;
w = HnswSearchLayer(base, q, ep, efConstruction, lc, index, procinfo, collation, m, true, skipElement);
w = HnswSearchLayer(base, &q, ep, efConstruction, lc, index, procinfo, collation, m, true, skipElement);
/* Elements being deleted or skipped can help with search */
/* but should be removed before selecting neighbors */

View File

@@ -1,608 +0,0 @@
#include "postgres.h"
#include <limits.h>
#include <math.h>
#include "catalog/pg_type.h"
#include "fmgr.h"
#include "intvec.h"
#include "lib/stringinfo.h"
#include "libpq/pqformat.h"
#include "utils/array.h"
#include "utils/builtins.h"
#include "utils/lsyscache.h"
/*
* Ensure same dimensions
*/
static inline void
CheckDims(IntVector * a, IntVector * b)
{
if (a->dim != b->dim)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("different intvec dimensions %d and %d", a->dim, b->dim)));
}
/*
* Ensure expected dimensions
*/
static inline void
CheckExpectedDim(int32 typmod, int dim)
{
if (typmod != -1 && typmod != dim)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("expected %d dimensions, not %d", typmod, dim)));
}
/*
* Ensure valid dimensions
*/
static inline void
CheckDim(int dim)
{
if (dim < 1)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("intvec must have at least 1 dimension")));
if (dim > INTVEC_MAX_DIM)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("intvec cannot have more than %d dimensions", INTVEC_MAX_DIM)));
}
/*
* Ensure element in range
*/
static inline void
CheckElement(long value)
{
if (value < SCHAR_MIN || value > SCHAR_MAX)
ereport(ERROR,
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
errmsg("value \"%ld\" is out of range for type intvec", value)));
}
/*
* Allocate and initialize a new int vector
*/
IntVector *
InitIntVector(int dim)
{
IntVector *result;
int size;
size = INTVEC_SIZE(dim);
result = (IntVector *) palloc0(size);
SET_VARSIZE(result, size);
result->dim = dim;
return result;
}
/*
* Check for whitespace, since array_isspace() is static
*/
static inline bool
intvec_isspace(char ch)
{
if (ch == ' ' ||
ch == '\t' ||
ch == '\n' ||
ch == '\r' ||
ch == '\v' ||
ch == '\f')
return true;
return false;
}
/*
* Convert textual representation to internal representation
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_in);
Datum
intvec_in(PG_FUNCTION_ARGS)
{
char *lit = PG_GETARG_CSTRING(0);
int32 typmod = PG_GETARG_INT32(2);
int8 x[INTVEC_MAX_DIM];
int dim = 0;
char *pt;
char *stringEnd;
IntVector *result;
char *litcopy = pstrdup(lit);
char *str = litcopy;
while (intvec_isspace(*str))
str++;
if (*str != '[')
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("malformed intvec literal: \"%s\"", lit),
errdetail("Vector contents must start with \"[\".")));
str++;
pt = strtok(str, ",");
stringEnd = pt;
while (pt != NULL && *stringEnd != ']')
{
long l;
if (dim == INTVEC_MAX_DIM)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("intvec cannot have more than %d dimensions", INTVEC_MAX_DIM)));
while (intvec_isspace(*pt))
pt++;
/* Check for empty string like float4in */
if (*pt == '\0')
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("invalid input syntax for type intvec: \"%s\"", lit)));
/* Use similar logic as int2vectorin */
errno = 0;
l = strtol(pt, &stringEnd, 10);
if (stringEnd == pt)
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("invalid input syntax for type intvec: \"%s\"", lit)));
if (errno == ERANGE || l < SCHAR_MIN || l > SCHAR_MAX)
ereport(ERROR,
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
errmsg("value \"%s\" is out of range for type intvec", pt)));
x[dim++] = l;
while (intvec_isspace(*stringEnd))
stringEnd++;
if (*stringEnd != '\0' && *stringEnd != ']')
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("invalid input syntax for type intvec: \"%s\"", lit)));
pt = strtok(NULL, ",");
}
if (stringEnd == NULL || *stringEnd != ']')
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("malformed intvec literal: \"%s\"", lit),
errdetail("Unexpected end of input.")));
stringEnd++;
/* Only whitespace is allowed after the closing brace */
while (intvec_isspace(*stringEnd))
stringEnd++;
if (*stringEnd != '\0')
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("malformed intvec literal: \"%s\"", lit),
errdetail("Junk after closing right brace.")));
/* Ensure no consecutive delimiters since strtok skips */
for (pt = lit + 1; *pt != '\0'; pt++)
{
if (pt[-1] == ',' && *pt == ',')
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("malformed intvec literal: \"%s\"", lit)));
}
if (dim < 1)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("intvec must have at least 1 dimension")));
pfree(litcopy);
CheckExpectedDim(typmod, dim);
result = InitIntVector(dim);
for (int i = 0; i < dim; i++)
result->x[i] = x[i];
PG_RETURN_POINTER(result);
}
/*
* Convert internal representation to textual representation
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_out);
Datum
intvec_out(PG_FUNCTION_ARGS)
{
IntVector *vector = PG_GETARG_INTVEC_P(0);
int dim = vector->dim;
char *buf;
char *ptr;
/*
* Need:
*
* dim * 4 bytes for elements (-128 to 127)
*
* dim - 1 bytes for separator
*
* 3 bytes for [, ], and \0
*/
buf = (char *) palloc(5 * dim + 2);
ptr = buf;
*ptr = '[';
ptr++;
for (int i = 0; i < dim; i++)
{
if (i > 0)
{
*ptr = ',';
ptr++;
}
#if PG_VERSION_NUM >= 140000
ptr += pg_ltoa(vector->x[i], ptr);
#else
pg_ltoa(vector->x[i], ptr);
while (*ptr != '\0')
ptr++;
#endif
}
*ptr = ']';
ptr++;
*ptr = '\0';
PG_FREE_IF_COPY(vector, 0);
PG_RETURN_CSTRING(buf);
}
/*
* Convert type modifier
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_typmod_in);
Datum
intvec_typmod_in(PG_FUNCTION_ARGS)
{
ArrayType *ta = PG_GETARG_ARRAYTYPE_P(0);
int32 *tl;
int n;
tl = ArrayGetIntegerTypmods(ta, &n);
if (n != 1)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("invalid type modifier")));
if (*tl < 1)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("dimensions for type intvec must be at least 1")));
if (*tl > INTVEC_MAX_DIM)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("dimensions for type intvec cannot exceed %d", INTVEC_MAX_DIM)));
PG_RETURN_INT32(*tl);
}
/*
* Convert external binary representation to internal representation
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_recv);
Datum
intvec_recv(PG_FUNCTION_ARGS)
{
StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
int32 typmod = PG_GETARG_INT32(2);
IntVector *result;
int16 dim;
int16 unused;
dim = pq_getmsgint(buf, sizeof(int16));
unused = pq_getmsgint(buf, sizeof(int16));
CheckDim(dim);
CheckExpectedDim(typmod, dim);
if (unused != 0)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("expected unused to be 0, not %d", unused)));
result = InitIntVector(dim);
for (int i = 0; i < dim; i++)
result->x[i] = pq_getmsgint(buf, sizeof(int8));
PG_RETURN_POINTER(result);
}
/*
* Convert internal representation to the external binary representation
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_send);
Datum
intvec_send(PG_FUNCTION_ARGS)
{
IntVector *vec = PG_GETARG_INTVEC_P(0);
StringInfoData buf;
pq_begintypsend(&buf);
pq_sendint(&buf, vec->dim, sizeof(int16));
pq_sendint(&buf, vec->unused, sizeof(int16));
for (int i = 0; i < vec->dim; i++)
pq_sendint8(&buf, vec->x[i]);
PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
}
/*
* Convert int vector to int vector
* This is needed to check the type modifier
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec);
Datum
intvec(PG_FUNCTION_ARGS)
{
IntVector *vec = PG_GETARG_INTVEC_P(0);
int32 typmod = PG_GETARG_INT32(1);
CheckExpectedDim(typmod, vec->dim);
PG_RETURN_POINTER(vec);
}
/*
* Convert array to intvec vector
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(array_to_intvec);
Datum
array_to_intvec(PG_FUNCTION_ARGS)
{
ArrayType *array = PG_GETARG_ARRAYTYPE_P(0);
int32 typmod = PG_GETARG_INT32(1);
Vector *result;
int16 typlen;
bool typbyval;
char typalign;
Datum *elemsp;
int nelemsp;
if (ARR_NDIM(array) > 1)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("array must be 1-D")));
if (ARR_HASNULL(array) && array_contains_nulls(array))
ereport(ERROR,
(errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
errmsg("array must not contain nulls")));
get_typlenbyvalalign(ARR_ELEMTYPE(array), &typlen, &typbyval, &typalign);
deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, NULL, &nelemsp);
CheckDim(nelemsp);
CheckExpectedDim(typmod, nelemsp);
result = InitVector(nelemsp);
if (ARR_ELEMTYPE(array) == INT4OID)
{
for (int i = 0; i < nelemsp; i++)
{
long l = DatumGetInt32(elemsp[i]);
CheckElement(l);
result->x[i] = l;
}
}
else
{
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("unsupported array type")));
}
/*
* Free allocation from deconstruct_array. Do not free individual elements
* when pass-by-reference since they point to original array.
*/
pfree(elemsp);
PG_RETURN_POINTER(result);
}
/*
* Get the L2 distance between int vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_l2_distance);
Datum
intvec_l2_distance(PG_FUNCTION_ARGS)
{
IntVector *a = PG_GETARG_INTVEC_P(0);
IntVector *b = PG_GETARG_INTVEC_P(1);
int8 *ax = a->x;
int8 *bx = b->x;
int distance = 0;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
{
int diff = ax[i] - bx[i];
distance += diff * diff;
}
PG_RETURN_FLOAT8(sqrt((double) distance));
}
/*
* Get the L2 squared distance between int vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_l2_squared_distance);
Datum
intvec_l2_squared_distance(PG_FUNCTION_ARGS)
{
IntVector *a = PG_GETARG_INTVEC_P(0);
IntVector *b = PG_GETARG_INTVEC_P(1);
int8 *ax = a->x;
int8 *bx = b->x;
int distance = 0;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
{
int diff = ax[i] - bx[i];
distance += diff * diff;
}
PG_RETURN_FLOAT8((double) distance);
}
/*
* Get the inner product of two int vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_inner_product);
Datum
intvec_inner_product(PG_FUNCTION_ARGS)
{
IntVector *a = PG_GETARG_INTVEC_P(0);
IntVector *b = PG_GETARG_INTVEC_P(1);
int8 *ax = a->x;
int8 *bx = b->x;
int distance = 0;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
distance += ax[i] * bx[i];
PG_RETURN_FLOAT8((double) distance);
}
/*
* Get the negative inner product of two int vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_negative_inner_product);
Datum
intvec_negative_inner_product(PG_FUNCTION_ARGS)
{
IntVector *a = PG_GETARG_INTVEC_P(0);
IntVector *b = PG_GETARG_INTVEC_P(1);
int8 *ax = a->x;
int8 *bx = b->x;
int distance = 0;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
distance += ax[i] * bx[i];
PG_RETURN_FLOAT8((double) -distance);
}
/*
* Get the cosine distance between two int vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_cosine_distance);
Datum
intvec_cosine_distance(PG_FUNCTION_ARGS)
{
IntVector *a = PG_GETARG_INTVEC_P(0);
IntVector *b = PG_GETARG_INTVEC_P(1);
int8 *ax = a->x;
int8 *bx = b->x;
int distance = 0;
int norma = 0;
int normb = 0;
double similarity;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
{
int8 axi = ax[i];
int8 bxi = bx[i];
distance += axi * bxi;
norma += axi * axi;
normb += bxi * bxi;
}
/* Use sqrt(a * b) over sqrt(a) * sqrt(b) */
similarity = (double) distance / sqrt((double) norma * (double) normb);
#ifdef _MSC_VER
/* /fp:fast may not propagate NaN */
if (isnan(similarity))
PG_RETURN_FLOAT8(NAN);
#endif
/* Keep in range */
if (similarity > 1)
similarity = 1;
else if (similarity < -1)
similarity = -1;
PG_RETURN_FLOAT8(1 - similarity);
}
/*
* Get the L1 distance between two int vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_l1_distance);
Datum
intvec_l1_distance(PG_FUNCTION_ARGS)
{
IntVector *a = PG_GETARG_INTVEC_P(0);
IntVector *b = PG_GETARG_INTVEC_P(1);
int8 *ax = a->x;
int8 *bx = b->x;
int distance = 0;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
distance += abs(ax[i] - bx[i]);
PG_RETURN_FLOAT8((double) distance);
}
/*
* Get the L2 norm of an int vector
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_l2_norm);
Datum
intvec_l2_norm(PG_FUNCTION_ARGS)
{
IntVector *a = PG_GETARG_INTVEC_P(0);
int8 *ax = a->x;
int norm = 0;
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
norm += ax[i] * ax[i];
PG_RETURN_FLOAT8(sqrt((double) norm));
}

View File

@@ -1,23 +0,0 @@
#ifndef INTVEC_H
#define INTVEC_H
#include "vector.h"
#define INTVEC_MAX_DIM VECTOR_MAX_DIM
#define INTVEC_SIZE(_dim) (offsetof(IntVector, x) + sizeof(int8)*(_dim))
#define DatumGetIntVector(x) ((IntVector *) PG_DETOAST_DATUM(x))
#define PG_GETARG_INTVEC_P(x) DatumGetIntVector(PG_GETARG_DATUM(x))
#define PG_RETURN_INTVEC_P(x) PG_RETURN_POINTER(x)
typedef struct IntVector
{
int32 vl_len_; /* varlena header (do not touch directly!) */
int16 dim; /* number of dimensions */
int16 unused;
int8 x[FLEXIBLE_ARRAY_MEMBER];
} IntVector;
IntVector *InitIntVector(int dim);
#endif

View File

@@ -57,7 +57,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
*/
if (buildstate->kmeansnormprocinfo != NULL)
{
if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value))
if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value, buildstate->normvec))
return;
}
@@ -105,7 +105,7 @@ SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
/* Add sample */
AddSample(values, buildstate);
AddSample(values, state);
/* Reset memory context */
MemoryContextSwitchTo(oldCtx);
@@ -153,7 +153,7 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
/* Normalize if needed */
if (buildstate->normprocinfo != NULL)
{
if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value))
if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec))
return;
}
@@ -356,6 +356,9 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
buildstate->centers = VectorArrayInit(buildstate->lists, buildstate->dimensions);
buildstate->listInfo = palloc(sizeof(ListInfo) * buildstate->lists);
/* Reuse for each tuple */
buildstate->normvec = InitVector(buildstate->dimensions);
buildstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
"Ivfflat build temporary context",
ALLOCSET_DEFAULT_SIZES);
@@ -377,6 +380,7 @@ FreeBuildState(IvfflatBuildState * buildstate)
{
VectorArrayFree(buildstate->centers);
pfree(buildstate->listInfo);
pfree(buildstate->normvec);
#ifdef IVFFLAT_KMEANS_DEBUG
pfree(buildstate->listSums);

View File

@@ -172,6 +172,7 @@ typedef struct IvfflatBuildState
VectorArray samples;
VectorArray centers;
ListInfo *listInfo;
Vector *normvec;
#ifdef IVFFLAT_KMEANS_DEBUG
double inertia;
@@ -266,7 +267,7 @@ void VectorArrayFree(VectorArray arr);
void PrintVectorArray(char *msg, VectorArray arr);
void IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers);
FmgrInfo *IvfflatOptionalProcInfo(Relation index, uint16 procnum);
bool IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value);
bool IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
int IvfflatGetLists(Relation index);
void IvfflatGetMetaPageInfo(Relation index, int *lists, int *dimensions);
void IvfflatUpdateList(Relation index, ListInfo listInfo, BlockNumber insertPage, BlockNumber originalInsertPage, BlockNumber startPage, ForkNumber forkNum);

View File

@@ -85,7 +85,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
if (normprocinfo != NULL)
{
if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value))
if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, NULL))
return;
}

View File

@@ -293,7 +293,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
/* Fine if normalization fails */
if (so->normprocinfo != NULL)
IvfflatNormValue(so->normprocinfo, so->collation, &value);
IvfflatNormValue(so->normprocinfo, so->collation, &value, NULL);
}
IvfflatBench("GetScanLists", GetScanLists(scan, value));

View File

@@ -75,14 +75,16 @@ IvfflatOptionalProcInfo(Relation index, uint16 procnum)
* if it's different than the original value
*/
bool
IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value)
IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
{
double norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
if (norm > 0)
{
Vector *v = DatumGetVector(*value);
Vector *result = InitVector(v->dim);
if (result == NULL)
result = InitVector(v->dim);
for (int i = 0; i < v->dim; i++)
result->x[i] = v->x[i] / norm;

View File

@@ -1,15 +1,15 @@
CREATE TABLE t (val vector(3), val2 intvec(3));
INSERT INTO t (val, val2) VALUES ('[0,0,0]', '[0,0,0]'), ('[1,2,3]', '[1,2,3]'), ('[1,1,1]', '[1,1,1]'), (NULL, NULL);
CREATE TABLE t2 (val vector(3), val2 intvec(3));
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val vector(3));
\copy t TO 'results/data.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/data.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;
val | val2
---------+---------
[0,0,0] | [0,0,0]
[1,1,1] | [1,1,1]
[1,2,3] | [1,2,3]
|
val
---------
[0,0,0]
[1,1,1]
[1,2,3]
(4 rows)
DROP TABLE t;

View File

@@ -104,105 +104,105 @@ SELECT vector_norm('[3e37,4e37]')::real;
5e+37
(1 row)
SELECT l2_distance('[0,0]'::vector, '[3,4]');
SELECT l2_distance('[0,0]', '[3,4]');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('[0,0]'::vector, '[0,1]');
SELECT l2_distance('[0,0]', '[0,1]');
l2_distance
-------------
1
(1 row)
SELECT l2_distance('[1,2]'::vector, '[3]');
SELECT l2_distance('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT l2_distance('[3e38]'::vector, '[-3e38]');
SELECT l2_distance('[3e38]', '[-3e38]');
l2_distance
-------------
Infinity
(1 row)
SELECT inner_product('[1,2]'::vector, '[3,4]');
SELECT inner_product('[1,2]', '[3,4]');
inner_product
---------------
11
(1 row)
SELECT inner_product('[1,2]'::vector, '[3]');
SELECT inner_product('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT inner_product('[3e38]'::vector, '[3e38]');
SELECT inner_product('[3e38]', '[3e38]');
inner_product
---------------
Infinity
(1 row)
SELECT cosine_distance('[1,2]'::vector, '[2,4]');
SELECT cosine_distance('[1,2]', '[2,4]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,2]'::vector, '[0,0]');
SELECT cosine_distance('[1,2]', '[0,0]');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('[1,1]'::vector, '[1,1]');
SELECT cosine_distance('[1,1]', '[1,1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,0]'::vector, '[0,2]');
SELECT cosine_distance('[1,0]', '[0,2]');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('[1,1]'::vector, '[-1,-1]');
SELECT cosine_distance('[1,1]', '[-1,-1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[1,2]'::vector, '[3]');
SELECT cosine_distance('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT cosine_distance('[1,1]'::vector, '[1.1,1.1]');
SELECT cosine_distance('[1,1]', '[1.1,1.1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,1]'::vector, '[-1.1,-1.1]');
SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[3e38]'::vector, '[3e38]');
SELECT cosine_distance('[3e38]', '[3e38]');
cosine_distance
-----------------
NaN
(1 row)
SELECT l1_distance('[0,0]'::vector, '[3,4]');
SELECT l1_distance('[0,0]', '[3,4]');
l1_distance
-------------
7
(1 row)
SELECT l1_distance('[0,0]'::vector, '[0,1]');
SELECT l1_distance('[0,0]', '[0,1]');
l1_distance
-------------
1
(1 row)
SELECT l1_distance('[1,2]'::vector, '[3]');
SELECT l1_distance('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT l1_distance('[3e38]'::vector, '[-3e38]');
SELECT l1_distance('[3e38]', '[-3e38]');
l1_distance
-------------
Infinity

View File

@@ -1,26 +0,0 @@
SET enable_seqscan = off;
CREATE TABLE t (val intvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val intvec_cosine_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
val
---------
[1,1,1]
[1,2,3]
[1,2,4]
(3 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
count
-------
3
(1 row)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::intvec)) t2;
count
-------
3
(1 row)
DROP TABLE t;

View File

@@ -1,21 +0,0 @@
SET enable_seqscan = off;
CREATE TABLE t (val intvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val intvec_ip_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
val
---------
[1,2,4]
[1,2,3]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::intvec)) t2;
count
-------
4
(1 row)
DROP TABLE t;

View File

@@ -1,33 +0,0 @@
SET enable_seqscan = off;
CREATE TABLE t (val intvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val intvec_l2_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,2,4]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::intvec)) t2;
count
-------
4
(1 row)
SELECT COUNT(*) FROM t;
count
-------
5
(1 row)
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
-----
(0 rows)
DROP TABLE t;

View File

@@ -12,11 +12,14 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector)) t2;
count
-------
4
(1 row)
SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
val
---------
[0,0,0]
[1,1,1]
[1,2,3]
[1,2,4]
(4 rows)
SELECT COUNT(*) FROM t;
count

View File

@@ -116,30 +116,8 @@ SELECT '[1, ,3]'::vector;
ERROR: invalid input syntax for type vector: "[1, ,3]"
LINE 1: SELECT '[1, ,3]'::vector;
^
SELECT '[1,2,3]'::vector(3);
vector
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::vector(2);
ERROR: expected 2 dimensions, not 3
SELECT '[1,2,3]'::vector(3, 2);
ERROR: invalid type modifier
LINE 1: SELECT '[1,2,3]'::vector(3, 2);
^
SELECT '[1,2,3]'::vector('a');
ERROR: invalid input syntax for type integer: "a"
LINE 1: SELECT '[1,2,3]'::vector('a');
^
SELECT '[1,2,3]'::vector(0);
ERROR: dimensions for type vector must be at least 1
LINE 1: SELECT '[1,2,3]'::vector(0);
^
SELECT '[1,2,3]'::vector(16001);
ERROR: dimensions for type vector cannot exceed 16000
LINE 1: SELECT '[1,2,3]'::vector(16001);
^
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
unnest
---------

View File

@@ -1,92 +0,0 @@
SELECT l2_distance('[0,0]'::intvec, '[3,4]');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('[0,0]'::intvec, '[0,1]');
l2_distance
-------------
1
(1 row)
SELECT l2_distance('[1,2]'::intvec, '[3]');
ERROR: different intvec dimensions 2 and 1
SELECT '[0,0]'::intvec <-> '[3,4]';
?column?
----------
5
(1 row)
SELECT inner_product('[1,2]'::intvec, '[3,4]');
inner_product
---------------
11
(1 row)
SELECT inner_product('[1,2]'::intvec, '[3]');
ERROR: different intvec dimensions 2 and 1
SELECT inner_product('[127]'::intvec, '[127]');
inner_product
---------------
16129
(1 row)
SELECT '[1,2]'::intvec <#> '[3,4]';
?column?
----------
-11
(1 row)
SELECT cosine_distance('[1,2]'::intvec, '[2,4]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,2]'::intvec, '[0,0]');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('[1,1]'::intvec, '[1,1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,0]'::intvec, '[0,2]');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('[1,1]'::intvec, '[-1,-1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[1,2]'::intvec, '[3]');
ERROR: different intvec dimensions 2 and 1
SELECT '[1,2]'::intvec <=> '[2,4]';
?column?
----------
0
(1 row)
SELECT l1_distance('[0,0]'::intvec, '[3,4]');
l1_distance
-------------
7
(1 row)
SELECT l1_distance('[0,0]'::intvec, '[0,1]');
l1_distance
-------------
1
(1 row)
SELECT l1_distance('[1,2]'::intvec, '[3]');
ERROR: different intvec dimensions 2 and 1

View File

@@ -1,119 +0,0 @@
SELECT '[1,2,3]'::intvec;
intvec
---------
[1,2,3]
(1 row)
SELECT '[-1,-2,-3]'::intvec;
intvec
------------
[-1,-2,-3]
(1 row)
SELECT ' [ 1, 2 , 3 ] '::intvec;
intvec
---------
[1,2,3]
(1 row)
SELECT '[1.23456]'::intvec;
ERROR: invalid input syntax for type intvec: "[1.23456]"
LINE 1: SELECT '[1.23456]'::intvec;
^
SELECT '[hello,1]'::intvec;
ERROR: invalid input syntax for type intvec: "[hello,1]"
LINE 1: SELECT '[hello,1]'::intvec;
^
SELECT '[127,-128]'::intvec;
intvec
------------
[127,-128]
(1 row)
SELECT '[128,-129]'::intvec;
ERROR: value "128" is out of range for type intvec
LINE 1: SELECT '[128,-129]'::intvec;
^
SELECT '[1,2,3'::intvec;
ERROR: malformed intvec literal: "[1,2,3"
LINE 1: SELECT '[1,2,3'::intvec;
^
DETAIL: Unexpected end of input.
SELECT '[1,2,3]9'::intvec;
ERROR: malformed intvec literal: "[1,2,3]9"
LINE 1: SELECT '[1,2,3]9'::intvec;
^
DETAIL: Junk after closing right brace.
SELECT '1,2,3'::intvec;
ERROR: malformed intvec literal: "1,2,3"
LINE 1: SELECT '1,2,3'::intvec;
^
DETAIL: Vector contents must start with "[".
SELECT ''::intvec;
ERROR: malformed intvec literal: ""
LINE 1: SELECT ''::intvec;
^
DETAIL: Vector contents must start with "[".
SELECT '['::intvec;
ERROR: malformed intvec literal: "["
LINE 1: SELECT '['::intvec;
^
DETAIL: Unexpected end of input.
SELECT '[,'::intvec;
ERROR: malformed intvec literal: "[,"
LINE 1: SELECT '[,'::intvec;
^
DETAIL: Unexpected end of input.
SELECT '[]'::intvec;
ERROR: intvec must have at least 1 dimension
LINE 1: SELECT '[]'::intvec;
^
SELECT '[1,]'::intvec;
ERROR: invalid input syntax for type intvec: "[1,]"
LINE 1: SELECT '[1,]'::intvec;
^
SELECT '[1a]'::intvec;
ERROR: invalid input syntax for type intvec: "[1a]"
LINE 1: SELECT '[1a]'::intvec;
^
SELECT '[1,,3]'::intvec;
ERROR: malformed intvec literal: "[1,,3]"
LINE 1: SELECT '[1,,3]'::intvec;
^
SELECT '[1, ,3]'::intvec;
ERROR: invalid input syntax for type intvec: "[1, ,3]"
LINE 1: SELECT '[1, ,3]'::intvec;
^
SELECT '[1,2,3]'::intvec(3);
intvec
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::intvec(2);
ERROR: expected 2 dimensions, not 3
SELECT '[1,2,3]'::intvec(3, 2);
ERROR: invalid type modifier
LINE 1: SELECT '[1,2,3]'::intvec(3, 2);
^
SELECT '[1,2,3]'::intvec('a');
ERROR: invalid input syntax for type integer: "a"
LINE 1: SELECT '[1,2,3]'::intvec('a');
^
SELECT '[1,2,3]'::intvec(0);
ERROR: dimensions for type intvec must be at least 1
LINE 1: SELECT '[1,2,3]'::intvec(0);
^
SELECT '[1,2,3]'::intvec(16001);
ERROR: dimensions for type intvec cannot exceed 16000
LINE 1: SELECT '[1,2,3]'::intvec(16001);
^
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::intvec[]);
unnest
---------
[1,2,3]
[4,5,6]
(2 rows)
SELECT '{"[1,2,3]"}'::intvec(2)[];
ERROR: expected 2 dimensions, not 3

View File

@@ -1,7 +1,7 @@
CREATE TABLE t (val vector(3), val2 intvec(3));
INSERT INTO t (val, val2) VALUES ('[0,0,0]', '[0,0,0]'), ('[1,2,3]', '[1,2,3]'), ('[1,1,1]', '[1,1,1]'), (NULL, NULL);
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val vector(3), val2 intvec(3));
CREATE TABLE t2 (val vector(3));
\copy t TO 'results/data.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/data.bin' WITH (FORMAT binary)

View File

@@ -24,29 +24,29 @@ SELECT vector_norm('[3,4]');
SELECT vector_norm('[0,1]');
SELECT vector_norm('[3e37,4e37]')::real;
SELECT l2_distance('[0,0]'::vector, '[3,4]');
SELECT l2_distance('[0,0]'::vector, '[0,1]');
SELECT l2_distance('[1,2]'::vector, '[3]');
SELECT l2_distance('[3e38]'::vector, '[-3e38]');
SELECT l2_distance('[0,0]', '[3,4]');
SELECT l2_distance('[0,0]', '[0,1]');
SELECT l2_distance('[1,2]', '[3]');
SELECT l2_distance('[3e38]', '[-3e38]');
SELECT inner_product('[1,2]'::vector, '[3,4]');
SELECT inner_product('[1,2]'::vector, '[3]');
SELECT inner_product('[3e38]'::vector, '[3e38]');
SELECT inner_product('[1,2]', '[3,4]');
SELECT inner_product('[1,2]', '[3]');
SELECT inner_product('[3e38]', '[3e38]');
SELECT cosine_distance('[1,2]'::vector, '[2,4]');
SELECT cosine_distance('[1,2]'::vector, '[0,0]');
SELECT cosine_distance('[1,1]'::vector, '[1,1]');
SELECT cosine_distance('[1,0]'::vector, '[0,2]');
SELECT cosine_distance('[1,1]'::vector, '[-1,-1]');
SELECT cosine_distance('[1,2]'::vector, '[3]');
SELECT cosine_distance('[1,1]'::vector, '[1.1,1.1]');
SELECT cosine_distance('[1,1]'::vector, '[-1.1,-1.1]');
SELECT cosine_distance('[3e38]'::vector, '[3e38]');
SELECT cosine_distance('[1,2]', '[2,4]');
SELECT cosine_distance('[1,2]', '[0,0]');
SELECT cosine_distance('[1,1]', '[1,1]');
SELECT cosine_distance('[1,0]', '[0,2]');
SELECT cosine_distance('[1,1]', '[-1,-1]');
SELECT cosine_distance('[1,2]', '[3]');
SELECT cosine_distance('[1,1]', '[1.1,1.1]');
SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
SELECT cosine_distance('[3e38]', '[3e38]');
SELECT l1_distance('[0,0]'::vector, '[3,4]');
SELECT l1_distance('[0,0]'::vector, '[0,1]');
SELECT l1_distance('[1,2]'::vector, '[3]');
SELECT l1_distance('[3e38]'::vector, '[-3e38]');
SELECT l1_distance('[0,0]', '[3,4]');
SELECT l1_distance('[0,0]', '[0,1]');
SELECT l1_distance('[1,2]', '[3]');
SELECT l1_distance('[3e38]', '[-3e38]');
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;

View File

@@ -1,13 +0,0 @@
SET enable_seqscan = off;
CREATE TABLE t (val intvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val intvec_cosine_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::intvec)) t2;
DROP TABLE t;

View File

@@ -1,12 +0,0 @@
SET enable_seqscan = off;
CREATE TABLE t (val intvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val intvec_ip_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::intvec)) t2;
DROP TABLE t;

View File

@@ -1,16 +0,0 @@
SET enable_seqscan = off;
CREATE TABLE t (val intvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val intvec_l2_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::intvec)) t2;
SELECT COUNT(*) FROM t;
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
DROP TABLE t;

View File

@@ -7,7 +7,7 @@ CREATE INDEX ON t USING hnsw (val vector_l2_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector)) t2;
SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
SELECT COUNT(*) FROM t;
TRUNCATE t;

View File

@@ -22,13 +22,7 @@ SELECT '[1,]'::vector;
SELECT '[1a]'::vector;
SELECT '[1,,3]'::vector;
SELECT '[1, ,3]'::vector;
SELECT '[1,2,3]'::vector(3);
SELECT '[1,2,3]'::vector(2);
SELECT '[1,2,3]'::vector(3, 2);
SELECT '[1,2,3]'::vector('a');
SELECT '[1,2,3]'::vector(0);
SELECT '[1,2,3]'::vector(16001);
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
SELECT '{"[1,2,3]"}'::vector(2)[];

View File

@@ -1,21 +0,0 @@
SELECT l2_distance('[0,0]'::intvec, '[3,4]');
SELECT l2_distance('[0,0]'::intvec, '[0,1]');
SELECT l2_distance('[1,2]'::intvec, '[3]');
SELECT '[0,0]'::intvec <-> '[3,4]';
SELECT inner_product('[1,2]'::intvec, '[3,4]');
SELECT inner_product('[1,2]'::intvec, '[3]');
SELECT inner_product('[127]'::intvec, '[127]');
SELECT '[1,2]'::intvec <#> '[3,4]';
SELECT cosine_distance('[1,2]'::intvec, '[2,4]');
SELECT cosine_distance('[1,2]'::intvec, '[0,0]');
SELECT cosine_distance('[1,1]'::intvec, '[1,1]');
SELECT cosine_distance('[1,0]'::intvec, '[0,2]');
SELECT cosine_distance('[1,1]'::intvec, '[-1,-1]');
SELECT cosine_distance('[1,2]'::intvec, '[3]');
SELECT '[1,2]'::intvec <=> '[2,4]';
SELECT l1_distance('[0,0]'::intvec, '[3,4]');
SELECT l1_distance('[0,0]'::intvec, '[0,1]');
SELECT l1_distance('[1,2]'::intvec, '[3]');

View File

@@ -1,28 +0,0 @@
SELECT '[1,2,3]'::intvec;
SELECT '[-1,-2,-3]'::intvec;
SELECT ' [ 1, 2 , 3 ] '::intvec;
SELECT '[1.23456]'::intvec;
SELECT '[hello,1]'::intvec;
SELECT '[127,-128]'::intvec;
SELECT '[128,-129]'::intvec;
SELECT '[1,2,3'::intvec;
SELECT '[1,2,3]9'::intvec;
SELECT '1,2,3'::intvec;
SELECT ''::intvec;
SELECT '['::intvec;
SELECT '[,'::intvec;
SELECT '[]'::intvec;
SELECT '[1,]'::intvec;
SELECT '[1a]'::intvec;
SELECT '[1,,3]'::intvec;
SELECT '[1, ,3]'::intvec;
SELECT '[1,2,3]'::intvec(3);
SELECT '[1,2,3]'::intvec(2);
SELECT '[1,2,3]'::intvec(3, 2);
SELECT '[1,2,3]'::intvec('a');
SELECT '[1,2,3]'::intvec(0);
SELECT '[1,2,3]'::intvec(16001);
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::intvec[]);
SELECT '{"[1,2,3]"}'::intvec(2)[];

View File

@@ -86,7 +86,7 @@ foreach (@queries)
push(@expected, $res);
}
test_recall(0.19, $limit, "before vacuum");
test_recall(0.20, $limit, "before vacuum");
test_recall(0.95, 100, "before vacuum");
# TODO Test concurrent inserts with vacuum

View File

@@ -1,132 +0,0 @@
use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More;
my $node;
my @queries = ();
my @expected;
my $limit = 20;
my $dim = 20;
my $array_sql = join(",", ('(random() * 255)::int - 128') x $dim);
sub test_recall
{
my ($min, $operator) = @_;
my $correct = 0;
my $total = 0;
my $explain = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator '$queries[0]' LIMIT $limit;
));
like($explain, qr/Index Scan/);
for my $i (0 .. $#queries)
{
my $actual = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SELECT i FROM tst ORDER BY v $operator '$queries[$i]' LIMIT $limit;
));
my @actual_ids = split("\n", $actual);
my %actual_set = map { $_ => 1 } @actual_ids;
my @expected_ids = split("\n", $expected[$i]);
foreach (@expected_ids)
{
if (exists($actual_set{$_}))
{
$correct++;
}
$total++;
}
}
cmp_ok($correct / $total, ">=", $min, $operator);
}
# Initialize node
$node = get_new_node('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v intvec($dim));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 10000) i;"
);
# Generate queries
for (1 .. 20)
{
my @r = ();
for (1 .. $dim)
{
push(@r, int(rand(256)) - 128);
}
push(@queries, "[" . join(",", @r) . "]");
}
# Check each index type
my @operators = ("<->", "<#>", "<=>");
my @opclasses = ("intvec_l2_ops", "intvec_ip_ops", "intvec_cosine_ops");
for my $i (0 .. $#operators)
{
my $operator = $operators[$i];
my $opclass = $opclasses[$i];
# Get exact results
@expected = ();
foreach (@queries)
{
my $res = $node->safe_psql("postgres", "SELECT i FROM tst ORDER BY v $operator '$_' LIMIT $limit;");
push(@expected, $res);
}
# Build index serially
$node->safe_psql("postgres", qq(
SET max_parallel_maintenance_workers = 0;
CREATE INDEX idx ON tst USING hnsw (v $opclass);
));
# Test approximate results
my $min = 0.99;
test_recall($min, $operator);
$node->safe_psql("postgres", "DROP INDEX idx;");
# Build index in parallel in memory
my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
SET client_min_messages = DEBUG;
SET min_parallel_table_scan_size = 1;
CREATE INDEX idx ON tst USING hnsw (v $opclass);
));
is($ret, 0, $stderr);
like($stderr, qr/using \d+ parallel workers/);
# Test approximate results
test_recall($min, $operator);
$node->safe_psql("postgres", "DROP INDEX idx;");
# Build index in parallel on disk
# Set parallel_workers on table to use workers with low maintenance_work_mem
($ret, $stdout, $stderr) = $node->psql("postgres", qq(
ALTER TABLE tst SET (parallel_workers = 2);
SET client_min_messages = DEBUG;
SET maintenance_work_mem = '4MB';
CREATE INDEX idx ON tst USING hnsw (v $opclass);
ALTER TABLE tst RESET (parallel_workers);
));
is($ret, 0, $stderr);
like($stderr, qr/using \d+ parallel workers/);
like($stderr, qr/hnsw graph no longer fits into maintenance_work_mem/);
$node->safe_psql("postgres", "DROP INDEX idx;");
}
done_testing();

View File

@@ -1,4 +1,4 @@
comment = 'vector data type and ivfflat and hnsw access methods'
default_version = '0.6.2'
default_version = '0.6.0'
module_pathname = '$libdir/vector'
relocatable = true